1 /* 2 * Copyright (c) 1982, 1986, 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94 39 * $FreeBSD: src/sys/kern/vfs_vnops.c,v 1.87.2.13 2002/12/29 18:19:53 dillon Exp $ 40 * $DragonFly: src/sys/kern/vfs_vnops.c,v 1.58 2008/06/28 17:59:49 dillon Exp $ 41 */ 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/fcntl.h> 46 #include <sys/file.h> 47 #include <sys/stat.h> 48 #include <sys/proc.h> 49 #include <sys/priv.h> 50 #include <sys/mount.h> 51 #include <sys/nlookup.h> 52 #include <sys/vnode.h> 53 #include <sys/buf.h> 54 #include <sys/filio.h> 55 #include <sys/ttycom.h> 56 #include <sys/conf.h> 57 #include <sys/sysctl.h> 58 #include <sys/syslog.h> 59 60 #include <sys/thread2.h> 61 #include <sys/mplock2.h> 62 63 static int vn_closefile (struct file *fp); 64 static int vn_ioctl (struct file *fp, u_long com, caddr_t data, 65 struct ucred *cred, struct sysmsg *msg); 66 static int vn_read (struct file *fp, struct uio *uio, 67 struct ucred *cred, int flags); 68 static int vn_kqfilter (struct file *fp, struct knote *kn); 69 static int vn_statfile (struct file *fp, struct stat *sb, struct ucred *cred); 70 static int vn_write (struct file *fp, struct uio *uio, 71 struct ucred *cred, int flags); 72 73 struct fileops vnode_fileops = { 74 .fo_read = vn_read, 75 .fo_write = vn_write, 76 .fo_ioctl = vn_ioctl, 77 .fo_kqfilter = vn_kqfilter, 78 .fo_stat = vn_statfile, 79 .fo_close = vn_closefile, 80 .fo_shutdown = nofo_shutdown 81 }; 82 83 /* 84 * Common code for vnode open operations. Check permissions, and call 85 * the VOP_NOPEN or VOP_NCREATE routine. 86 * 87 * The caller is responsible for setting up nd with nlookup_init() and 88 * for cleaning it up with nlookup_done(), whether we return an error 89 * or not. 90 * 91 * On success nd->nl_open_vp will hold a referenced and, if requested, 92 * locked vnode. A locked vnode is requested via NLC_LOCKVP. If fp 93 * is non-NULL the vnode will be installed in the file pointer. 94 * 95 * NOTE: The vnode is referenced just once on return whether or not it 96 * is also installed in the file pointer. 97 */ 98 int 99 vn_open(struct nlookupdata *nd, struct file *fp, int fmode, int cmode) 100 { 101 struct vnode *vp; 102 struct ucred *cred = nd->nl_cred; 103 struct vattr vat; 104 struct vattr *vap = &vat; 105 int error; 106 u_int flags; 107 108 /* 109 * Certain combinations are illegal 110 */ 111 if ((fmode & (FWRITE | O_TRUNC)) == O_TRUNC) 112 return(EACCES); 113 114 /* 115 * Lookup the path and create or obtain the vnode. After a 116 * successful lookup a locked nd->nl_nch will be returned. 117 * 118 * The result of this section should be a locked vnode. 119 * 120 * XXX with only a little work we should be able to avoid locking 121 * the vnode if FWRITE, O_CREAT, and O_TRUNC are *not* set. 122 */ 123 nd->nl_flags |= NLC_OPEN; 124 if (fmode & O_APPEND) 125 nd->nl_flags |= NLC_APPEND; 126 if (fmode & O_TRUNC) 127 nd->nl_flags |= NLC_TRUNCATE; 128 if (fmode & FREAD) 129 nd->nl_flags |= NLC_READ; 130 if (fmode & FWRITE) 131 nd->nl_flags |= NLC_WRITE; 132 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 133 nd->nl_flags |= NLC_FOLLOW; 134 135 if (fmode & O_CREAT) { 136 /* 137 * CONDITIONAL CREATE FILE CASE 138 * 139 * Setting NLC_CREATE causes a negative hit to store 140 * the negative hit ncp and not return an error. Then 141 * nc_error or nc_vp may be checked to see if the ncp 142 * represents a negative hit. NLC_CREATE also requires 143 * write permission on the governing directory or EPERM 144 * is returned. 145 */ 146 nd->nl_flags |= NLC_CREATE; 147 nd->nl_flags |= NLC_REFDVP; 148 bwillinode(1); 149 error = nlookup(nd); 150 } else { 151 /* 152 * NORMAL OPEN FILE CASE 153 */ 154 error = nlookup(nd); 155 } 156 157 if (error) 158 return (error); 159 160 /* 161 * split case to allow us to re-resolve and retry the ncp in case 162 * we get ESTALE. 163 */ 164 again: 165 if (fmode & O_CREAT) { 166 if (nd->nl_nch.ncp->nc_vp == NULL) { 167 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 168 return (error); 169 VATTR_NULL(vap); 170 vap->va_type = VREG; 171 vap->va_mode = cmode; 172 if (fmode & O_EXCL) 173 vap->va_vaflags |= VA_EXCLUSIVE; 174 error = VOP_NCREATE(&nd->nl_nch, nd->nl_dvp, &vp, 175 nd->nl_cred, vap); 176 if (error) 177 return (error); 178 fmode &= ~O_TRUNC; 179 /* locked vnode is returned */ 180 } else { 181 if (fmode & O_EXCL) { 182 error = EEXIST; 183 } else { 184 error = cache_vget(&nd->nl_nch, cred, 185 LK_EXCLUSIVE, &vp); 186 } 187 if (error) 188 return (error); 189 fmode &= ~O_CREAT; 190 } 191 } else { 192 error = cache_vget(&nd->nl_nch, cred, LK_EXCLUSIVE, &vp); 193 if (error) 194 return (error); 195 } 196 197 /* 198 * We have a locked vnode and ncp now. Note that the ncp will 199 * be cleaned up by the caller if nd->nl_nch is left intact. 200 */ 201 if (vp->v_type == VLNK) { 202 error = EMLINK; 203 goto bad; 204 } 205 if (vp->v_type == VSOCK) { 206 error = EOPNOTSUPP; 207 goto bad; 208 } 209 if ((fmode & O_CREAT) == 0) { 210 if (fmode & (FWRITE | O_TRUNC)) { 211 if (vp->v_type == VDIR) { 212 error = EISDIR; 213 goto bad; 214 } 215 error = vn_writechk(vp, &nd->nl_nch); 216 if (error) { 217 /* 218 * Special stale handling, re-resolve the 219 * vnode. 220 */ 221 if (error == ESTALE) { 222 vput(vp); 223 vp = NULL; 224 cache_setunresolved(&nd->nl_nch); 225 error = cache_resolve(&nd->nl_nch, cred); 226 if (error == 0) 227 goto again; 228 } 229 goto bad; 230 } 231 } 232 } 233 if (fmode & O_TRUNC) { 234 vn_unlock(vp); /* XXX */ 235 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 236 VATTR_NULL(vap); 237 vap->va_size = 0; 238 error = VOP_SETATTR(vp, vap, cred); 239 if (error) 240 goto bad; 241 } 242 243 /* 244 * Set or clear VNSWAPCACHE on the vp based on nd->nl_nch.ncp->nc_flag. 245 * These particular bits a tracked all the way from the root. 246 * 247 * NOTE: Might not work properly on NFS servers due to the 248 * disconnected namecache. 249 */ 250 flags = nd->nl_nch.ncp->nc_flag; 251 if ((flags & (NCF_UF_CACHE | NCF_UF_PCACHE)) && 252 (flags & (NCF_SF_NOCACHE | NCF_SF_PNOCACHE)) == 0) { 253 vsetflags(vp, VSWAPCACHE); 254 } else { 255 vclrflags(vp, VSWAPCACHE); 256 } 257 258 /* 259 * Setup the fp so VOP_OPEN can override it. No descriptor has been 260 * associated with the fp yet so we own it clean. 261 * 262 * f_nchandle inherits nl_nch. This used to be necessary only for 263 * directories but now we do it unconditionally so f*() ops 264 * such as fchmod() can access the actual namespace that was 265 * used to open the file. 266 */ 267 if (fp) { 268 if (nd->nl_flags & NLC_APPENDONLY) 269 fmode |= FAPPENDONLY; 270 fp->f_nchandle = nd->nl_nch; 271 cache_zero(&nd->nl_nch); 272 cache_unlock(&fp->f_nchandle); 273 } 274 275 /* 276 * Get rid of nl_nch. vn_open does not return it (it returns the 277 * vnode or the file pointer). Note: we can't leave nl_nch locked 278 * through the VOP_OPEN anyway since the VOP_OPEN may block, e.g. 279 * on /dev/ttyd0 280 */ 281 if (nd->nl_nch.ncp) 282 cache_put(&nd->nl_nch); 283 284 error = VOP_OPEN(vp, fmode, cred, fp); 285 if (error) { 286 /* 287 * setting f_ops to &badfileops will prevent the descriptor 288 * code from trying to close and release the vnode, since 289 * the open failed we do not want to call close. 290 */ 291 if (fp) { 292 fp->f_data = NULL; 293 fp->f_ops = &badfileops; 294 } 295 goto bad; 296 } 297 298 #if 0 299 /* 300 * Assert that VREG files have been setup for vmio. 301 */ 302 KASSERT(vp->v_type != VREG || vp->v_object != NULL, 303 ("vn_open: regular file was not VMIO enabled!")); 304 #endif 305 306 /* 307 * Return the vnode. XXX needs some cleaning up. The vnode is 308 * only returned in the fp == NULL case. 309 */ 310 if (fp == NULL) { 311 nd->nl_open_vp = vp; 312 nd->nl_vp_fmode = fmode; 313 if ((nd->nl_flags & NLC_LOCKVP) == 0) 314 vn_unlock(vp); 315 } else { 316 vput(vp); 317 } 318 return (0); 319 bad: 320 if (vp) 321 vput(vp); 322 return (error); 323 } 324 325 int 326 vn_opendisk(const char *devname, int fmode, struct vnode **vpp) 327 { 328 struct vnode *vp; 329 int error; 330 331 if (strncmp(devname, "/dev/", 5) == 0) 332 devname += 5; 333 if ((vp = getsynthvnode(devname)) == NULL) { 334 error = ENODEV; 335 } else { 336 error = VOP_OPEN(vp, fmode, proc0.p_ucred, NULL); 337 vn_unlock(vp); 338 if (error) { 339 vrele(vp); 340 vp = NULL; 341 } 342 } 343 *vpp = vp; 344 return (error); 345 } 346 347 /* 348 * Check for write permissions on the specified vnode. nch may be NULL. 349 */ 350 int 351 vn_writechk(struct vnode *vp, struct nchandle *nch) 352 { 353 /* 354 * If there's shared text associated with 355 * the vnode, try to free it up once. If 356 * we fail, we can't allow writing. 357 */ 358 if (vp->v_flag & VTEXT) 359 return (ETXTBSY); 360 361 /* 362 * If the vnode represents a regular file, check the mount 363 * point via the nch. This may be a different mount point 364 * then the one embedded in the vnode (e.g. nullfs). 365 * 366 * We can still write to non-regular files (e.g. devices) 367 * via read-only mounts. 368 */ 369 if (nch && nch->ncp && vp->v_type == VREG) 370 return (ncp_writechk(nch)); 371 return (0); 372 } 373 374 /* 375 * Check whether the underlying mount is read-only. The mount point 376 * referenced by the namecache may be different from the mount point 377 * used by the underlying vnode in the case of NULLFS, so a separate 378 * check is needed. 379 */ 380 int 381 ncp_writechk(struct nchandle *nch) 382 { 383 if (nch->mount && (nch->mount->mnt_flag & MNT_RDONLY)) 384 return (EROFS); 385 return(0); 386 } 387 388 /* 389 * Vnode close call 390 * 391 * MPSAFE 392 */ 393 int 394 vn_close(struct vnode *vp, int flags) 395 { 396 int error; 397 398 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 399 if (error == 0) { 400 error = VOP_CLOSE(vp, flags); 401 vn_unlock(vp); 402 } 403 vrele(vp); 404 return (error); 405 } 406 407 /* 408 * Sequential heuristic. 409 * 410 * MPSAFE (f_seqcount and f_nextoff are allowed to race) 411 */ 412 static __inline 413 int 414 sequential_heuristic(struct uio *uio, struct file *fp) 415 { 416 /* 417 * Sequential heuristic - detect sequential operation 418 * 419 * NOTE: SMP: We allow f_seqcount updates to race. 420 */ 421 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 422 uio->uio_offset == fp->f_nextoff) { 423 int tmpseq = fp->f_seqcount; 424 425 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE; 426 if (tmpseq > IO_SEQMAX) 427 tmpseq = IO_SEQMAX; 428 fp->f_seqcount = tmpseq; 429 return(fp->f_seqcount << IO_SEQSHIFT); 430 } 431 432 /* 433 * Not sequential, quick draw-down of seqcount 434 * 435 * NOTE: SMP: We allow f_seqcount updates to race. 436 */ 437 if (fp->f_seqcount > 1) 438 fp->f_seqcount = 1; 439 else 440 fp->f_seqcount = 0; 441 return(0); 442 } 443 444 /* 445 * get - lock and return the f_offset field. 446 * set - set and unlock the f_offset field. 447 * 448 * These routines serve the dual purpose of serializing access to the 449 * f_offset field (at least on i386) and guaranteeing operational integrity 450 * when multiple read()ers and write()ers are present on the same fp. 451 * 452 * MPSAFE 453 */ 454 static __inline off_t 455 vn_get_fpf_offset(struct file *fp) 456 { 457 u_int flags; 458 u_int nflags; 459 460 /* 461 * Shortcut critical path. 462 */ 463 flags = fp->f_flag & ~FOFFSETLOCK; 464 if (atomic_cmpset_int(&fp->f_flag, flags, flags | FOFFSETLOCK)) 465 return(fp->f_offset); 466 467 /* 468 * The hard way 469 */ 470 for (;;) { 471 flags = fp->f_flag; 472 if (flags & FOFFSETLOCK) { 473 nflags = flags | FOFFSETWAKE; 474 tsleep_interlock(&fp->f_flag, 0); 475 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) 476 tsleep(&fp->f_flag, PINTERLOCKED, "fpoff", 0); 477 } else { 478 nflags = flags | FOFFSETLOCK; 479 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) 480 break; 481 } 482 } 483 return(fp->f_offset); 484 } 485 486 /* 487 * MPSAFE 488 */ 489 static __inline void 490 vn_set_fpf_offset(struct file *fp, off_t offset) 491 { 492 u_int flags; 493 u_int nflags; 494 495 /* 496 * We hold the lock so we can set the offset without interference. 497 */ 498 fp->f_offset = offset; 499 500 /* 501 * Normal release is already a reasonably critical path. 502 */ 503 for (;;) { 504 flags = fp->f_flag; 505 nflags = flags & ~(FOFFSETLOCK | FOFFSETWAKE); 506 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) { 507 if (flags & FOFFSETWAKE) 508 wakeup(&fp->f_flag); 509 break; 510 } 511 } 512 } 513 514 /* 515 * MPSAFE 516 */ 517 static __inline off_t 518 vn_poll_fpf_offset(struct file *fp) 519 { 520 #if defined(__x86_64__) || !defined(SMP) 521 return(fp->f_offset); 522 #else 523 off_t off = vn_get_fpf_offset(fp); 524 vn_set_fpf_offset(fp, off); 525 return(off); 526 #endif 527 } 528 529 /* 530 * Package up an I/O request on a vnode into a uio and do it. 531 * 532 * MPSAFE 533 */ 534 int 535 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 536 off_t offset, enum uio_seg segflg, int ioflg, 537 struct ucred *cred, int *aresid) 538 { 539 struct uio auio; 540 struct iovec aiov; 541 struct ccms_lock ccms_lock; 542 int error; 543 544 if ((ioflg & IO_NODELOCKED) == 0) 545 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 546 auio.uio_iov = &aiov; 547 auio.uio_iovcnt = 1; 548 aiov.iov_base = base; 549 aiov.iov_len = len; 550 auio.uio_resid = len; 551 auio.uio_offset = offset; 552 auio.uio_segflg = segflg; 553 auio.uio_rw = rw; 554 auio.uio_td = curthread; 555 ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, &auio); 556 if (rw == UIO_READ) { 557 error = VOP_READ(vp, &auio, ioflg, cred); 558 } else { 559 error = VOP_WRITE(vp, &auio, ioflg, cred); 560 } 561 ccms_lock_put(&vp->v_ccms, &ccms_lock); 562 if (aresid) 563 *aresid = auio.uio_resid; 564 else 565 if (auio.uio_resid && error == 0) 566 error = EIO; 567 if ((ioflg & IO_NODELOCKED) == 0) 568 vn_unlock(vp); 569 return (error); 570 } 571 572 /* 573 * Package up an I/O request on a vnode into a uio and do it. The I/O 574 * request is split up into smaller chunks and we try to avoid saturating 575 * the buffer cache while potentially holding a vnode locked, so we 576 * check bwillwrite() before calling vn_rdwr(). We also call lwkt_user_yield() 577 * to give other processes a chance to lock the vnode (either other processes 578 * core'ing the same binary, or unrelated processes scanning the directory). 579 * 580 * MPSAFE 581 */ 582 int 583 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 584 off_t offset, enum uio_seg segflg, int ioflg, 585 struct ucred *cred, int *aresid) 586 { 587 int error = 0; 588 589 do { 590 int chunk; 591 592 /* 593 * Force `offset' to a multiple of MAXBSIZE except possibly 594 * for the first chunk, so that filesystems only need to 595 * write full blocks except possibly for the first and last 596 * chunks. 597 */ 598 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 599 600 if (chunk > len) 601 chunk = len; 602 if (vp->v_type == VREG) { 603 switch(rw) { 604 case UIO_READ: 605 bwillread(chunk); 606 break; 607 case UIO_WRITE: 608 bwillwrite(chunk); 609 break; 610 } 611 } 612 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 613 ioflg, cred, aresid); 614 len -= chunk; /* aresid calc already includes length */ 615 if (error) 616 break; 617 offset += chunk; 618 base += chunk; 619 lwkt_user_yield(); 620 } while (len); 621 if (aresid) 622 *aresid += len; 623 return (error); 624 } 625 626 /* 627 * File pointers can no longer get ripped up by revoke so 628 * we don't need to lock access to the vp. 629 * 630 * f_offset updates are not guaranteed against multiple readers 631 * 632 * MPSAFE 633 */ 634 static int 635 vn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 636 { 637 struct ccms_lock ccms_lock; 638 struct vnode *vp; 639 int error, ioflag; 640 641 KASSERT(uio->uio_td == curthread, 642 ("uio_td %p is not td %p", uio->uio_td, curthread)); 643 vp = (struct vnode *)fp->f_data; 644 645 ioflag = 0; 646 if (flags & O_FRNONBLOCKING) { 647 ioflag |= (IO_NDELAY | IO_NRDELAY); 648 } else if (flags & O_FBLOCKING) { 649 /* ioflag &= ~IO_NDELAY; */ 650 } else if (flags & O_FNONBLOCKING) { 651 ioflag |= IO_NDELAY; 652 } else if (fp->f_flag & FNONBLOCK) { 653 ioflag |= IO_NDELAY; 654 } 655 if (flags & O_FBUFFERED) { 656 /* ioflag &= ~IO_DIRECT; */ 657 } else if (flags & O_FUNBUFFERED) { 658 ioflag |= IO_DIRECT; 659 } else if (fp->f_flag & O_DIRECT) { 660 ioflag |= IO_DIRECT; 661 } 662 if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0) 663 uio->uio_offset = vn_get_fpf_offset(fp); 664 vn_lock(vp, LK_SHARED | LK_RETRY); 665 ioflag |= sequential_heuristic(uio, fp); 666 667 ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, uio); 668 error = VOP_READ(vp, uio, ioflag, cred); 669 ccms_lock_put(&vp->v_ccms, &ccms_lock); 670 fp->f_nextoff = uio->uio_offset; 671 vn_unlock(vp); 672 if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0) 673 vn_set_fpf_offset(fp, uio->uio_offset); 674 return (error); 675 } 676 677 /* 678 * MPSAFE 679 */ 680 static int 681 vn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 682 { 683 struct ccms_lock ccms_lock; 684 struct vnode *vp; 685 int error, ioflag; 686 687 KASSERT(uio->uio_td == curthread, 688 ("uio_td %p is not p %p", uio->uio_td, curthread)); 689 vp = (struct vnode *)fp->f_data; 690 691 ioflag = IO_UNIT; 692 if (vp->v_type == VREG && 693 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 694 ioflag |= IO_APPEND; 695 } 696 697 if (flags & O_FBLOCKING) { 698 /* ioflag &= ~IO_NDELAY; */ 699 } else if (flags & O_FNONBLOCKING) { 700 ioflag |= IO_NDELAY; 701 } else if (fp->f_flag & FNONBLOCK) { 702 ioflag |= IO_NDELAY; 703 } 704 if (flags & O_FBUFFERED) { 705 /* ioflag &= ~IO_DIRECT; */ 706 } else if (flags & O_FUNBUFFERED) { 707 ioflag |= IO_DIRECT; 708 } else if (fp->f_flag & O_DIRECT) { 709 ioflag |= IO_DIRECT; 710 } 711 if (flags & O_FASYNCWRITE) { 712 /* ioflag &= ~IO_SYNC; */ 713 } else if (flags & O_FSYNCWRITE) { 714 ioflag |= IO_SYNC; 715 } else if (fp->f_flag & O_FSYNC) { 716 ioflag |= IO_SYNC; 717 } 718 719 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 720 ioflag |= IO_SYNC; 721 if ((flags & O_FOFFSET) == 0) 722 uio->uio_offset = vn_get_fpf_offset(fp); 723 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 724 ioflag |= sequential_heuristic(uio, fp); 725 ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, uio); 726 error = VOP_WRITE(vp, uio, ioflag, cred); 727 ccms_lock_put(&vp->v_ccms, &ccms_lock); 728 fp->f_nextoff = uio->uio_offset; 729 vn_unlock(vp); 730 if ((flags & O_FOFFSET) == 0) 731 vn_set_fpf_offset(fp, uio->uio_offset); 732 return (error); 733 } 734 735 /* 736 * MPSAFE 737 */ 738 static int 739 vn_statfile(struct file *fp, struct stat *sb, struct ucred *cred) 740 { 741 struct vnode *vp; 742 int error; 743 744 vp = (struct vnode *)fp->f_data; 745 error = vn_stat(vp, sb, cred); 746 return (error); 747 } 748 749 /* 750 * MPSAFE 751 */ 752 int 753 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *cred) 754 { 755 struct vattr vattr; 756 struct vattr *vap; 757 int error; 758 u_short mode; 759 cdev_t dev; 760 761 vap = &vattr; 762 error = VOP_GETATTR(vp, vap); 763 if (error) 764 return (error); 765 766 /* 767 * Zero the spare stat fields 768 */ 769 sb->st_lspare = 0; 770 sb->st_qspare1 = 0; 771 sb->st_qspare2 = 0; 772 773 /* 774 * Copy from vattr table 775 */ 776 if (vap->va_fsid != VNOVAL) 777 sb->st_dev = vap->va_fsid; 778 else 779 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 780 sb->st_ino = vap->va_fileid; 781 mode = vap->va_mode; 782 switch (vap->va_type) { 783 case VREG: 784 mode |= S_IFREG; 785 break; 786 case VDATABASE: 787 mode |= S_IFDB; 788 break; 789 case VDIR: 790 mode |= S_IFDIR; 791 break; 792 case VBLK: 793 mode |= S_IFBLK; 794 break; 795 case VCHR: 796 mode |= S_IFCHR; 797 break; 798 case VLNK: 799 mode |= S_IFLNK; 800 /* This is a cosmetic change, symlinks do not have a mode. */ 801 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) 802 sb->st_mode &= ~ACCESSPERMS; /* 0000 */ 803 else 804 sb->st_mode |= ACCESSPERMS; /* 0777 */ 805 break; 806 case VSOCK: 807 mode |= S_IFSOCK; 808 break; 809 case VFIFO: 810 mode |= S_IFIFO; 811 break; 812 default: 813 return (EBADF); 814 } 815 sb->st_mode = mode; 816 if (vap->va_nlink > (nlink_t)-1) 817 sb->st_nlink = (nlink_t)-1; 818 else 819 sb->st_nlink = vap->va_nlink; 820 sb->st_uid = vap->va_uid; 821 sb->st_gid = vap->va_gid; 822 sb->st_rdev = dev2udev(vp->v_rdev); 823 sb->st_size = vap->va_size; 824 sb->st_atimespec = vap->va_atime; 825 sb->st_mtimespec = vap->va_mtime; 826 sb->st_ctimespec = vap->va_ctime; 827 828 /* 829 * A VCHR and VBLK device may track the last access and last modified 830 * time independantly of the filesystem. This is particularly true 831 * because device read and write calls may bypass the filesystem. 832 */ 833 if (vp->v_type == VCHR || vp->v_type == VBLK) { 834 dev = vp->v_rdev; 835 if (dev != NULL) { 836 if (dev->si_lastread) { 837 sb->st_atimespec.tv_sec = dev->si_lastread; 838 sb->st_atimespec.tv_nsec = 0; 839 } 840 if (dev->si_lastwrite) { 841 sb->st_atimespec.tv_sec = dev->si_lastwrite; 842 sb->st_atimespec.tv_nsec = 0; 843 } 844 } 845 } 846 847 /* 848 * According to www.opengroup.org, the meaning of st_blksize is 849 * "a filesystem-specific preferred I/O block size for this 850 * object. In some filesystem types, this may vary from file 851 * to file" 852 * Default to PAGE_SIZE after much discussion. 853 */ 854 855 if (vap->va_type == VREG) { 856 sb->st_blksize = vap->va_blocksize; 857 } else if (vn_isdisk(vp, NULL)) { 858 /* 859 * XXX this is broken. If the device is not yet open (aka 860 * stat() call, aka v_rdev == NULL), how are we supposed 861 * to get a valid block size out of it? 862 */ 863 dev = vp->v_rdev; 864 865 sb->st_blksize = dev->si_bsize_best; 866 if (sb->st_blksize < dev->si_bsize_phys) 867 sb->st_blksize = dev->si_bsize_phys; 868 if (sb->st_blksize < BLKDEV_IOSIZE) 869 sb->st_blksize = BLKDEV_IOSIZE; 870 } else { 871 sb->st_blksize = PAGE_SIZE; 872 } 873 874 sb->st_flags = vap->va_flags; 875 876 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0); 877 if (error) 878 sb->st_gen = 0; 879 else 880 sb->st_gen = (u_int32_t)vap->va_gen; 881 882 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 883 return (0); 884 } 885 886 /* 887 * MPALMOSTSAFE - acquires mplock 888 */ 889 static int 890 vn_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *ucred, 891 struct sysmsg *msg) 892 { 893 struct vnode *vp = ((struct vnode *)fp->f_data); 894 struct vnode *ovp; 895 struct vattr vattr; 896 int error; 897 off_t size; 898 899 switch (vp->v_type) { 900 case VREG: 901 case VDIR: 902 if (com == FIONREAD) { 903 error = VOP_GETATTR(vp, &vattr); 904 if (error) 905 break; 906 size = vattr.va_size; 907 if ((vp->v_flag & VNOTSEEKABLE) == 0) 908 size -= vn_poll_fpf_offset(fp); 909 if (size > 0x7FFFFFFF) 910 size = 0x7FFFFFFF; 911 *(int *)data = size; 912 error = 0; 913 break; 914 } 915 if (com == FIOASYNC) { /* XXX */ 916 error = 0; /* XXX */ 917 break; 918 } 919 /* fall into ... */ 920 default: 921 #if 0 922 return (ENOTTY); 923 #endif 924 case VFIFO: 925 case VCHR: 926 case VBLK: 927 if (com == FIODTYPE) { 928 if (vp->v_type != VCHR && vp->v_type != VBLK) { 929 error = ENOTTY; 930 break; 931 } 932 *(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK; 933 error = 0; 934 break; 935 } 936 error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred, msg); 937 if (error == 0 && com == TIOCSCTTY) { 938 struct proc *p = curthread->td_proc; 939 struct session *sess; 940 941 if (p == NULL) { 942 error = ENOTTY; 943 break; 944 } 945 946 get_mplock(); 947 sess = p->p_session; 948 /* Do nothing if reassigning same control tty */ 949 if (sess->s_ttyvp == vp) { 950 error = 0; 951 rel_mplock(); 952 break; 953 } 954 955 /* Get rid of reference to old control tty */ 956 ovp = sess->s_ttyvp; 957 vref(vp); 958 sess->s_ttyvp = vp; 959 if (ovp) 960 vrele(ovp); 961 rel_mplock(); 962 } 963 break; 964 } 965 return (error); 966 } 967 968 /* 969 * Check that the vnode is still valid, and if so 970 * acquire requested lock. 971 */ 972 int 973 #ifndef DEBUG_LOCKS 974 vn_lock(struct vnode *vp, int flags) 975 #else 976 debug_vn_lock(struct vnode *vp, int flags, const char *filename, int line) 977 #endif 978 { 979 int error; 980 981 do { 982 #ifdef DEBUG_LOCKS 983 vp->filename = filename; 984 vp->line = line; 985 error = debuglockmgr(&vp->v_lock, flags, 986 "vn_lock", filename, line); 987 #else 988 error = lockmgr(&vp->v_lock, flags); 989 #endif 990 if (error == 0) 991 break; 992 } while (flags & LK_RETRY); 993 994 /* 995 * Because we (had better!) have a ref on the vnode, once it 996 * goes to VRECLAIMED state it will not be recycled until all 997 * refs go away. So we can just check the flag. 998 */ 999 if (error == 0 && (vp->v_flag & VRECLAIMED)) { 1000 lockmgr(&vp->v_lock, LK_RELEASE); 1001 error = ENOENT; 1002 } 1003 return (error); 1004 } 1005 1006 #ifdef DEBUG_VN_UNLOCK 1007 1008 void 1009 debug_vn_unlock(struct vnode *vp, const char *filename, int line) 1010 { 1011 kprintf("vn_unlock from %s:%d\n", filename, line); 1012 lockmgr(&vp->v_lock, LK_RELEASE); 1013 } 1014 1015 #else 1016 1017 void 1018 vn_unlock(struct vnode *vp) 1019 { 1020 lockmgr(&vp->v_lock, LK_RELEASE); 1021 } 1022 1023 #endif 1024 1025 /* 1026 * MPSAFE 1027 */ 1028 int 1029 vn_islocked(struct vnode *vp) 1030 { 1031 return (lockstatus(&vp->v_lock, curthread)); 1032 } 1033 1034 /* 1035 * Return the lock status of a vnode and unlock the vnode 1036 * if we owned the lock. This is not a boolean, if the 1037 * caller cares what the lock status is the caller must 1038 * check the various possible values. 1039 * 1040 * This only unlocks exclusive locks held by the caller, 1041 * it will NOT unlock shared locks (there is no way to 1042 * tell who the shared lock belongs to). 1043 * 1044 * MPSAFE 1045 */ 1046 int 1047 vn_islocked_unlock(struct vnode *vp) 1048 { 1049 int vpls; 1050 1051 vpls = lockstatus(&vp->v_lock, curthread); 1052 if (vpls == LK_EXCLUSIVE) 1053 lockmgr(&vp->v_lock, LK_RELEASE); 1054 return(vpls); 1055 } 1056 1057 /* 1058 * Restore a vnode lock that we previously released via 1059 * vn_islocked_unlock(). This is a NOP if we did not 1060 * own the original lock. 1061 * 1062 * MPSAFE 1063 */ 1064 void 1065 vn_islocked_relock(struct vnode *vp, int vpls) 1066 { 1067 int error; 1068 1069 if (vpls == LK_EXCLUSIVE) 1070 error = lockmgr(&vp->v_lock, vpls); 1071 } 1072 1073 /* 1074 * MPSAFE 1075 */ 1076 static int 1077 vn_closefile(struct file *fp) 1078 { 1079 int error; 1080 1081 fp->f_ops = &badfileops; 1082 error = vn_close(((struct vnode *)fp->f_data), fp->f_flag); 1083 return (error); 1084 } 1085 1086 /* 1087 * MPSAFE 1088 */ 1089 static int 1090 vn_kqfilter(struct file *fp, struct knote *kn) 1091 { 1092 int error; 1093 1094 error = VOP_KQFILTER(((struct vnode *)fp->f_data), kn); 1095 return (error); 1096 } 1097